Face Detection

The face detection node finds faces in images and video frames and, optionally, computes face embeddings (vectors) that can be used for recognition and clustering. Video is processed by scanning frames at a configurable rate.

Kind

facedetect

Applies to

Image, Video

Inputs

None

Output keys

face_count (Integer), facedetect_flag (String); face bounding boxes + embedding vectors are persisted as detections

Requirements

InspireFace native runtime and its model pack. CPU works; a GPU-enabled pack accelerates video/large batches. Model files require local storage and memory.

Persists to

detection table (bulk upsert) + asset_node_result ledger

Configuration

FacedetectNodeOptions:

Option Meaning

inspirefacePackPath

Path to the InspireFace model pack

capabilities

Which InspireFace capabilities to enable (detection, recognition, …)

videoChopRate

How often to sample frames from a video

videoScaleSize

Frame scale used before detection

minFaceHeightFactor

Minimum face size (as a fraction of frame height) to keep

faceClusterMinimum / faceClusterEPS

Clustering parameters for grouping embeddings into persons

Use Cases

  • Face search & recognition — match a query face against embeddings across the library.

  • Person clustering — group appearances of the same person via the embedding clusters.

  • Content moderation / review — flag assets that contain people.

  • Feed Face Description with detected faces.